{"id":"W2288957490","doi":"10.20355/c5kg6p","title":"Fairness of Standardized Assessments: Discrepancy between Provincial and Territorial Results","year":2016,"lang":"en","type":"article","venue":"Journal of Contemporary Issues in Education","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Psychology; Section (typography); Standardized test; Regional science; Political science; Sociology; Mathematics education; Business; Advertising","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00677743,0.0001017997,0.0004269094,0.0004208985,0.00003825588,0.00008496574,0.0003141197,0.0000675991,0.0000627208],"category_scores_gemma":[0.002199217,0.0000578171,0.0000692503,0.0002697785,0.0000837313,0.001637198,0.00004450885,0.0001053008,0.000003044313],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001210224,"about_ca_system_score_gemma":0.002273844,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008415749,"about_ca_topic_score_gemma":0.00003739541,"domain_scores_codex":[0.996083,0.0004502118,0.001735923,0.0001785037,0.001452813,0.00009954823],"domain_scores_gemma":[0.9962929,0.0007934629,0.001588858,0.0002425414,0.0009961888,0.00008606995],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001278486,0.0002946203,0.6471732,0.00001747467,0.00003549341,0.000002102275,0.002810349,0.000001279893,0.0008419693,0.001110467,0.01417879,0.3322558],"study_design_scores_gemma":[0.00699584,0.001269234,0.6212274,0.0008147982,0.00002942155,0.000009074044,0.005940535,0.00002268761,0.003710876,0.04522949,0.3145181,0.0002325115],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9822248,0.000634745,0.001690055,0.00815837,0.003262751,0.0002502982,0.00005887483,0.000004287105,0.00371583],"genre_scores_gemma":[0.9959414,0.0001091497,0.001561781,0.00002275349,0.00142715,0.000004574767,0.000004698551,0.000005686928,0.0009228237],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3320232,"threshold_uncertainty_score":0.4033701,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1183209692239725,"score_gpt":0.5121441527236013,"score_spread":0.3938231834996287,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}